Characterization of weak convergence of Birkhoff sums for Gibbs-Markov maps (Q616509): Difference between revisions

This nice article investigates limit theorems for Birkhoff sums of locally Lipschitzian functions under the iteration of so-called Gibbs-Markov maps. Aaronson and Denker have given sufficient conditions to have limit theorems in this setting. The author shows that these conditions are also necessary so that there is no exotic limit theorem for Gibbs-Markov maps. His proofs, valid under very weak regularity assumptions on the observable, involve weak perturbation theory and interpolation spaces. More precisely, the author considers on a bounded metric space \(J(X,d)\) endowed with a probability measure \(m\), a mixing probability-preserving map \(T\) \((X\to X)\) which satisfies some assumptions (summarized as ``Gibbs-Markov'') which prescribe in particular that \(T\) is strictly expanding and Markovian with respect to some measurable partition of \(X\) (a finite part of which playing a predominant role). Then, under some weak assumption on a locally Lipschitz real function \(f\) on \(X\), the author establishes the following result on Birkhoff sums \(S_n f:= \sum^{n-1}_{k=0} f\circ T^k\): -- either \(f= u- u_0T+ c\in L^2\), and then \(u\) is bounded and \((S_n f-nc)\to Z-Z'\) (in law, with \(Z\), \(Z'\) independent and having the law of \(u\)); -- or \(f\in L^2\) and \((S_n f-n\int f\,dm)\to{\mathcal N}(0,\sigma^2)\) (with some explicit variance \(\sigma^2\)); -- or else \(f\not\in L^2\), and \(S_nf\) has the same (known) asymptotic behaviour as the sum of \(n\) i.i.d. random variable (distributed as \(f\)). In the \(L^2\) case, the author obtains also necessary and sufficient conditions to control the speed of convergence in the central limit theorem.